New APPS

Art, Politics, Philosophy, Science. Mostly Philosophy.

recent posts

about

    • Yet another terrible AI idea

      People make snap judgments about those they see the first time – mentally categorizing someone as friendly, threatening, trustworthy, etc.  Most of us know that those impressions are idiosyncratic, and suffused with cultural biases along race, gender and other lines.  So obviously I know what you’re thinking… we need an AI that do that, right?  At least that’s what this new PNAS paper seems to think (h/t Nico Osaka for the link).  The authors start right in with the significance:

      “We quickly and irresistibly form impressions of what other people are like based solely on how their faces look. These impressions have real-life consequences ranging from hiring decisions to sentencing decisions. We model and visualize the perceptual bases of facial impressions in the most comprehensive fashion to date, producing photorealistic models of 34 perceived social and physical attributes (e.g., trustworthiness and age). These models leverage and demonstrate the utility of deep learning in face evaluation, allowing for 1) generation of an infinite number of faces that vary along these perceived attribute dimensions, 2) manipulation of any face photograph along these dimensions, and 3) prediction of the impressions any face image may evoke in the general (mostly White, North American) population”

      Let’s maybe think for a minute, yes?  Because we know that people make these impressions on unsound bases!

      First, adversarial networks are already able to produce fake faces that are indistinguishable from real ones.  Those fake faces can now be manipulated to appear more or less trustworthy, hostile, friendly, etc.  When you make fake political ads, for example, that’s going to be useful.  Already 6 years ago, one “Melvin Redick of Harrisburg, Pa., a friendly-looking American with a backward baseball cap and a young daughter, posted on Facebook a link to a brand-new website,” saying on June 8, 2016 that “these guys show hidden truth about Hillary Clinton, George Soros and other leaders of the US. Visit #DCLeaks website. It’s really interesting!” Of course, both Melvin Redick and the site he pointed to were complete fabrications by the Russians.  Now we can make Melvin look trustworthy, and Clinton less so.

      Second, the ability to manipulate existing face photos is a disaster-in-waiting.  Again, we saw crude efforts with this before – making Obama appear darker than he is, for example.  But here news photos could be altered to make Vladimir Putin appear trustworthy, or Mr. Rogers untrustworthy.  This goes nowhere good, especially when combined with deepfake technology that already takes people out of their contexts and puts them in other ones (disproportionately, so far, women are pasted into porn videos, but the Russians recently tried to produce a deepfake of Zelensky surrendering.  Fortunately that one was done sloppily).

      Third, and I think this one is possibly the scariest, what about scanning images to see whether someone will be assessed as trustworthy?  AI-based hiring is already under-regulated!  Now employers will run your photo through the software and make customer-service hiring decisions based on who customers will perceive as trustworthy.  What could go wrong?

      All of this of course assumes that this sort of software actually works.  The history of physiognomic AI, which uses all sorts of supposedly objective cues to determine personality and which is basically complete (usually racist) bunk suggests that the science is probably not as good as the article acts like.  So maybe we’re lucky and this algorithm does not actually work as advertised.  Of course, the fact that AI software is garbage doesn't preclude its being used to make people's lives miserable.  Just consider the bizarre case of VibraImage.

      But don’t worry.  The PNAS authors are aware of ethics, noting that “the framework developed here adds significantly to the ethical concerns that already enshroud image manipulation software:”

      “Our model can induce (perceived) changes within the individual’s face itself and may be difficult to detect when applied subtly enough. We argue that such methods (as well as their implementations and supporting data) should be made transparent from the start, such that the community can develop robust detection and defense protocols to accompany the technology, as they have done, for example, in developing highly accurate image forensics techniques to detect synthetic faces generated by SG2. More generally, to the extent that improper use of the image manipulation techniques described here is not covered by existing defamation law, it is appropriate to consider ways to limit use of these technologies through regulatory frameworks proposed in the broader context of face-recognition technologies.” 

      Yes, the very effective American regulation of privacy does inspire confidence!  Also, “There is also potential for our data and models to perpetuate the biases they measure, which are first impressions of the population under study and have no necessary correspondence to the actual identities, attitudes, or competencies of people whom the images resemble or depict.”

      Do you think?  As Luke Stark put it, facial recognition is the “plutonium of AI:” very dangerous and with very few legitimate uses.  This algorithm belongs in the same category, and should similarly be regulated like nuclear waste.  For example, as Ari Waldman and Mary Anne Franks have written, one of the problems with deepfakes is that the fake version gets out there on the internet, and it is nearly impossible to make it go away (if you even know about it).  Forensic software gets there too late, and those without resources aren’t going to be able to deploy it anyway.  Lawsuits are even less useful, since they're time-consuming and expensive to pursue, and lots of defendants won't be jurisdictionally available or have pockets deep enough to make the chase worth it.  In other words, not everybody is going to be able to defend themselves like Zelensky, who both warned about deepfakes and was able to produce video of himself not surrendering.  In the meantime, faked and shocking things generally get diffused faster and further than real news.  After all, “engagement” is the business model of social media.  Further, to the extent that people stay inside filter bubbles (like Fox News), they may never see the forensic corrections, and they probably won’t believe the real one is real, even if they do.

      And as for reinforcing existing biases, Safiya Noble already wrote a whole book on how algorithms that guess what you’re probably thinking about someone can do just that.

    • Copyright Office gets this right…

      in refusing to grant copyright registration to an AI creation.  I suspect this one to be litigated for a while, since the person who has been trying to get protection for the picture has declared limiting copyright to human authors as something that would be unconstitutional (I also think it would be pretty entertaining to watch somebody try to float that argument in front of the current Supreme Court).  A good article on why this sort of thing is going to be a problem, and an interesting way of parsing law's traditional 'mental state' requirement, is here.

    • AI Colonialism

      First read this piece by Abeba Birhane, who warns about neocolonial exploitation of people in Africa by AI and big tech.

      Then read this detailed account of content moderation for Facebook in Kenya and abuse of the workers involved.

    • IRS to start using Facial Recogntion Technology

      If you want to use their website; WaPo has the story here.  But it's one of those public/private partnerships where data leaks and hacks and thefts happen.  To their credit, the Post went to Joy Buolamwini, whose work proved that facial recognition systems work best on white men and worst on Black women.  But even a perfectly functioning system is frightening.  First, it would unquestionably worsen the divide between those who have good Internet and those who don't, making convenient access to tax records contingent on having a sufficient income and computing skills.  Also, of course, facial recognition is bad – the potential for misuse is so great, and the record is so permanent, that Woody Hartzog and Evan Selinger argue it ought to be legally impossible to consent to. (for an overview of the debate, see Selinger and Leong here).

      One of the biggest problems with data is that it gets leaked and hacked, of course, but another big problem is that companies sell it to pretty much whoever arrives with cash.  The company handling IRS facial recognition claims they'll turn it over to law enforcement, but the Post says there's no federal law proscribing what they can do with it.  And they're switching companies and authentication strategies because of a massive data breach at Equifax a few years ago. So its not like nobody has ever heard of a data breach.

      Oh, and ID.me, the company getting the contract, totally wants to sell you stuff:

      "But advertising is a key part of ID.me’s operation, too. People who sign up on ID.me’s website are asked if they want to subscribe to “offers and discounts” from the company’s online storefront, which links to special deals for veterans, students and first responders. Consumer marketing accounts for 10 percent of the company’s revenue."

      What could possibly go wrong? Well, if you look up the ID.me privacy policy, you discover that most of the usual things can go wrong.  For example, they don't police 3rd party use of the data, which they encourage you to opt-in to:

      "To avoid any confusion, Users should understand that, while we own and operate the Service and Website, we do not own or operate websites owned and operated by third parties who may avail themselves of the ID.me Service (collectively referred to hereafter as the “Third-Party Websites”). This Privacy Policy is intended to inform Users about our collection, use, storage, and disclosure, destruction and disposal of information that we collect or record in the course of providing the Website and the ID.me Service. Please note, we are not responsible for the privacy practices of Third-Party Websites and they are under no obligation to comply with this Privacy Policy. Before visiting Third-Party Websites, and before providing the User’s ID.me or any other information to any party that operates or advertises on Third-Party Websites, Users should review the privacy policy and practices of that website to determine how information collected from Users will be handled. Please further note, depending on a User’s particular interaction with us (e.g., Users who solely navigate the Website versus Users who create an account and use the ID.me Service at Third-Party Websites), different portions of this policy may apply to Users at different times."

      Also, they reserve the right to change their privacy policy at any time, and it's your job to read it frequently to see:

      "If we decide to change this Privacy Policy, we will post those changes to this page so that you are aware of what information we collect, how we use it, and under what circumstances, if any, we disclose it. We reserve the right to modify this Privacy Policy at any time, so please review it frequently. If we make material changes to this policy, we will notify you here, by email, or by means of notice on our home page."

      That's item 1 on the policy.  Nothing else matters.  This is typical corporate privacy boilerplate that lets them do whatever they want with your facial biometric information.  Good job IRS!

       

       

    • SCOTUS and Social Murder

      The SCOTUS decision yesterday striking down OSHA’s vaccine mandate is based on some of the most sophomoric reasoning the Court has issued in a long time.  And I am aware of what Court I’m talking about.  The gist of the argument is that OSHA is only authorized to enact safety rules that protect someone’s at their place of occupation.  But this is a public health rule because Covid also occurs outside the workplace, ergo etc.

      But of course work is one of the main places that you can get Covid, as Justin Feldman documents (he also shows that the predominance of workplace transmission helps to explain the disproportionate impact on non-white folks).  The fact that vaccination also protects you outside of work is nice but not the point.  I have a ladder at home.  I don’t know the OSHA rules, but I bet there’s some covering the construction and use of ladders at work.  If those rules cause ladder manufacturers to make a safer product, that also protects me at home.  But it’s a little hard to explain how that standard doesn’t meet the statutory mandate of protecting people who use ladders in their occupation (the dissent cites several more such examples).   What’s wrong with positive externalities?

      The Court opines:

      “It is telling that OSHA, in its half century of existence, has never before adopted a broad public health regulation of this kind—addressing a threat that is untethered, in any causal sense, from the workplace.”

      Well, duh.  We haven’t had a global pandemic like Covid during the existence of OSHA!  In the meantime, if you read court opinions very often, you learn to expect documentation of bold factual assertions like that one.  But there is no footnote explaining how there is no causal relation between the threat of Covid and the workplace.  That’s because a credible such footnote cannot be written.  As the dissent points out, “because the disease spreads in shared indoor spaces, it presents heightened dangers in most workplaces,” citing OSHA’s documentation of the risks and reminding that majority that Courts are supposed to be deferential in cases like this.  Congress even allocated money to OSHA  to address workplace hazards (dissent, p. 8).  In short,

      “The agency backed up its conclusions with hundreds of reports of workplace COVID–19 outbreaks—not just in cheek-by-jowl settings like factory assembly lines, but in retail stores, restaurants, medical facilities, construction areas, and standard offices.” (dissent, p. 9)

      We also know that SCOTUS doesn’t even believe its own rhetoric about workplace risk: the justices are all vaccinated, all but Gorsuch wore masks to oral arguments on this case (prompting Sotomayor to participate from her chambers), and court policy is that arguing attorneys have to take a Covid test the day before, and argue remotely if positive.  Attorneys are also supposed to wear KN95 masks when in the Courtroom except when actually speaking.  One of the attorneys arguing against the mandate even had to appear remotely because he had Covid!  So workplace safety is apparently a thing that SCOTUS has heard of – it’s just not one they deem fit to extend to workers who have less control over their environment.

      In the meantime, Gorsuch took the time to write a concurrence tediously saying that states might have authority for public health, and that the nondelegation doctrine “ensures democratic accountability by preventing Congress from intentionally delegating its legislative powers to unelected officials.”  Perhaps now is the time to remember that SCOTUS is unelected, and seems to enjoy its own antidemocratic powers quite a bit: this the Court that ordered the Biden administration to reinstate the Remain in Mexico policy, even though that’s foreign policy, traditionally the province of the democratically elected executive (remember, the Court kept trying to greenlight Trump’s border wall with the fake border Caravan emergency, even though Congress specifically withheld funding for it).  This is also the same Justice Gorsuch who was appointed by the minoritarian Senate at the invitation of Donald Trump because Mitch McConnell refused to consider the nomination of the person who was democratically-elected president at the time of the vacancy. (Gorsuch also pontificates about the “major questions doctrine,” which is supposed intervene when an “agency may seek to exploit some gap, ambiguity, or doubtful expression in Congress’s statutes to assume responsibilities far beyond its initial assignment.”  But since the Court made no effort to prove that a vaccination mandate would not improve workplace safety and instead tries to show that the mandate improved safety everywhere, this rhetoric should be filed under the ‘I’m going to cite myself in anti-regulatory rulings in the future” dept).

      There is one bit of hope in the opinion, in this paragraph:

      “That is not to say OSHA lacks authority to regulate occupation-specific risks related to COVID–19. Where the virus poses a special danger because of the particular features of an employee’s job or workplace, targeted regulations are plainly permissible. We do not doubt, for example, that OSHA could regulate researchers who work with the COVID–19 virus. So too could OSHA regulate risks associated with working in particularly crowded or cramped environments. But the danger present in such workplaces differs in both degree and kind from the everyday risk of contracting COVID–19 that all face. OSHA’s indiscriminate approach fails to account for this crucial distinction—between occupational risk and risk more generally—and accordingly the mandate takes on the character of a general public health measure, rather than an “occupational safety or health standard.” 29 U. S. C. §655(b) (emphasis added).” (slip op, p. 7)

      The Biden administration should immediately institute revised standards mandating vaccination in places with disproportionately high Covid rates.  There’s been research on that; as CNBC reports of the study:

      “The top five occupations that had higher than a 50% mortality rate increase during the pandemic include cooks, line workers in warehouses, agricultural workers, bakers and construction laborers.”

      Feldman links to some other high risk groups.  But the Biden administration needs to immediately call the Court’s bluff.   Will SCOTUS reverse itself here and go full-on Lochner and declare that the baking profession is unregulable?

      Marx had lots of words for how the capitalist class treated the lives of workers as disposable.  Engels had the better expression: “social murder.”  How many workers did the right-wing majority in SCOTUS kill yesterday?  “OSHA estimated that in six months the emergency standard would save over 6,500 lives and prevent over 250,000 hospitalizations” (dissent, p. 11), and that number was derived before Omicron emerged. As the dissent sums it up:

      “Underlying everything else in this dispute is a single, simple question: Who decides how much protection, and of what kind, American workers need from COVID–19? An agency with expertise in workplace health and safety, acting as Congress and the President authorized? Or a court, lacking any knowledge of how to safeguard workplaces, and insulated from responsibility for any damage it causes?”

      There’s definitely a separation of powers problem emerging, but it’s not the one the Court’s conservatives want you to think about.

    • Can Datasets be Governed Like Public Utilities?

      By Gordon Hull

      Machine Learning (ML) applications learn by repetition.  That is, they come to recognize what, say, a chair looks like, by seeing lots of images of chairs that have been correctly labeled as such.  Since the machine is trying to figure out a pattern or set of characteristics that distinguish chairs from other objects, the more chairs it sees, the better it will perform at its task.  This collection of labeled chairs is the algorithm’s “training data.”   There are notorious problems with bias in training data due to underrepresentation of certain groups.  For example, one study found that datasets designed to train ML to recognize objects performed poorly in developing countries, most likely due to underrepresentation of images from those places; when combined with pictures labeled in English, and the fact that standard Western commodity forms of household objects might look very different in the developing world, the ML was stumped.  Most famously in this country, facial recognition software performs best at identifying white men and worst at identifying Black women.  ImageNet, which is widely used for object recognition purposes, employees a hierarchy of labels that include calling a child wearing sunglasses a “failure, loser, non-starter, unsuccessful person” but also differentiates between assistant and associate professors.  Despite these and many other problems, massive datasets are essential for training ML applications.

      For this reason, datasets have been called “infrastructural” for machine learning, defined as follows:

      “Infrastructure is characterized, we argue, by a set of core features: it is embedded into, and acts as the foundation, for other tools and technologies; when working as intended for a particular community, it tends to seep into the background and become incorporated into routines; the invisibility of infrastructure, however, is situated – what is natural or taken for granted from one perspective may be highly visible or jarring from another; though frequently naturalized, infrastructure is built, and thus inherently contextual, situated, and shaped by specific aims.”

      If AI is cars and buses and trains that do what we want, the datasets it trains on shape the roads and paths where those roads go, provide their material basis, become thereby incorporated into higher level routines like search, and tend to disappear into the background when not actively used.  But just like other examples of infrastructure – say, the bridges over the Long Island Freeway – infrastructure can embed priorities and affordances.  In this sense, dataset infrastructures have a politics, and serve as platforms on which applications are built.

      (more…)

    • For the Record…

      This is not what critical race theory says:

      "Critical race theory, Guelzo says, is a subset of critical theory that began with Immanuel Kant in the 1790s. It was a response to — and rejection of — the principles of the Enlightenment and the Age of Reason on which the American republic was founded. Kant believed that “reason was inadequate to give shape to our lives” and so he set about “developing a theory of being critical of reason,” Guelzo says."

      The linked piece is a podcast called "WTH is critical race theory? How a philosophy that inspired Marxism, Nazism, and Jim Crow is making its way into our schools, and what we can do."

      That's a right-wing columnist in the Washington Post bringing disgrace and ridicule to his paper for publishing this drivel.  Also, it's apparently far too easy for old white dudes to get tenure at Princeton, because they apparently don't have to read Kant before talking about him.  You don't even have to know that Marx wrote about 100 years before critical race theory.

      No, seriously.  What did that clown do to get tenure?  I only ask because conservatives claim that woke liberals are destroying academic standards with their social justice blah blah blah.

      Please nobody ever complain again about a leftist takeover of the academy. 

    • AI for Social Good: Calling out Amazon’s Lies to Congress

      (story via Julia Angwin) You might remember that Amazon solemnly swore to Congress that they did not artificially elevate their own products in search results.  Except that they do.  Adrianne Jeffries and Leon Yin of The Markup used a machine learning algorithm to predict product placement in search results:

      “We found that knowing only whether a product was an Amazon brand or exclusive could predict in seven out of every 10 cases whether Amazon would place it first in search results. These listings are not visibly marked as “sponsored” and they are part of a grid that Amazon identifies as “search results” in the site’s source code.”

      Nothing else was anywhere close to as predictive.  Amazon products routinely get rated higher than better rated and better-selling competitors.  And it isn’t just for obvious Amazon products like “Amazon Basics.”  It turns out that a lot of products turn out to be Amazon:

      “Using public records from the U.S. Patent and Trademark Office and Amazon’s own statements, we identified more than 150 brands registered by or owned by Amazon. These include both brands with an obvious connection, such as Amazon Basics and Amazon Commercial, and those that are generally known to be owned by the company, including Kindle and Zappos. But they also include dozens more, such as Happy Belly, Daily Ritual, and Society New York, where the connection to the company is not obvious. Those are in addition to the estimated hundreds of third-party brands that are exclusive to the site.”

      Consumers are (understandably) totally confused about who makes the products they see.  Actual independent sellers are terrified to complain, because they’re worried about retaliation from Amazon.  Those who want to elevate their products can of course pay (one of the companies listed in the article was paying $10,000 a month, another $40,000 – not small change!), though even that might not get them listed above the Amazon products.  All of this is, not surprisingly, a potential antitrust problem:

      “Bill Baer, a former assistant attorney general in charge of the antitrust division of the U.S. Department of Justice and former director of the Bureau of Competition at the FTC, said if consumers expect Amazon’s product search results to be neutral, but they are not, and the site is essentially a monopoly, that could be a violation of the FTC Act of 1914, which prohibits unfair competition and unfair or deceptive practices in commerce, or the U.S. Sherman Antirust Act, which prohibits monopolies from using their market power to harm competition.  If basically you’ve got somebody with market power that is restraining competition both in terms of site access or where things appear on the site,” he said, “that is potentially problematic.”

      It’s also a problem because it suggests that Amazon executives lied to Congress.  Congress, you’ll recall, is completely dysfunctional and polarized on all topics except one: they all hate big tech.  Hence the “social good” part of the AI: the House Judiciary committee just sent a very threatening letter, which opens as follows:

      “We write in response to recent, credible reporting that directly contradicts the sworn testimony and representations of Amazon’s top executives—including former CEO Jeffrey Bezos—to the Committee about their company’s business practices during our investigation last Congress. At best, this reporting confirms that Amazon’s representatives misled the Committee. At worst, it demonstrates that they may have lied to Congress in possible violation of federal criminal law.  In light of the serious nature of this matter, we are providing you with a final opportunity to provide exculpatory evidence to corroborate the prior testimony and statements on behalf of Amazon to the Committee. We strongly encourage you to make use of this opportunity to correct the record and provide the Committee with sworn, truthful, and accurate responses to this request as we consider whether a referral of this matter to the Department of Justice for criminal investigation is appropriate.”

    • Facebook as a Hostile Foreign Power

      This piece, on Facebook's behaving more like an autocratic, hostile state than a large company, is worth the read.  Here's an excerpt:

      "Perhaps Americans have become so cynical that they have given up on defending their freedom from surveillance, manipulation, and exploitation. But if Russia or China were taking the exact same actions to undermine democracy [that FB is], Americans would surely feel differently. Seeing Facebook as a hostile foreign power could force people to acknowledge what they’re participating in, and what they’re giving up, when they log in. In the end it doesn’t really matter what Facebook is; it matters what Facebook is doing."

    • Charles Mills (1951-2021)

      I was both saddened and stunned this morning to read of the passing of Charles Mills.  I first met him at a SPEP years ago; I was having lunch at some random sandwich shop with friends.  He knew one of us, and asked if he could join.  Nevermind that we were all junior.  I managed to find him at a few later conferences, to join a meal or even just to say hi.  He was generous, warm and wickedly funny when he wanted to be.  And the stunning clarity with which he could call out ideal philosophy and other systems of domination is like nothing else I’ve read.

      Daily Nous has more, and links to a really touching remembrance by Liam Kofi Bright.